A Complaint Text Classification Model Based on Character-Level Convolutional Network
Xuesong Tong, Bin Wu, Shuyang Wang, Jinna Lv · 2018
With the increase of demand for service quality, a growing number of people are expressing their complaints on the Web for services from different businesses. The correct classification of complaint reasons can substantially improve the quality of business service. Existing methods for text classification used on various datasets are mushrooming, however analysis on complaint texts is rare in current literature. There are still great challenges to classify complaint texts. On the one hand, complaint texts contain obvious negative sentiments which are useless for complaint text classification. On the other hand, complaint texts have more semantic and grammatical errors caused by negative emotions especially in Chinese, which enhances the difficulty of modeling. In response to the challenge, we propose a novel complaint text classification model based on character-level convolutional network. First, we employ a Negative Elements Removal (NER)module to denoise complaint texts. Second, in order to reduce effects of semantic and grammatical errors, a character-based convolutional network for complaint texts is proposed. Experiments demonstrate that our model can achieve state-of-the-art results on Chinese and English complaint texts compared with traditional methods and deep learning methods.